Ease of Teaching Tactics
-
I think there are two competing approaches with regard to thinking (and arguing) about what voting systems to prefer over what other voting systems on account of their characteristics related to tactics.
I'm not sure where the boundary between tactics and strategy is. I would welcome an explanation about that. For the purposes of the rest of this message, I will use "tactic" to mean the decision process (not necessarily formal) to take into account the voter's affinities toward the candidacies, the voter's estimate of where the other voters stand on the candidacies, and of course what voting system is in play, to decide the voter's vote.
In my opinion, our decisions as advocates should respect a right of every voter to equal power to what every other voter has, to influence the electoral outcome.
With regard to rating voting systems in a way that takes into account how the systems interact with tactics, I think one popular approach is to prefer systems about which no student of the field has publicly admitted to thinking of any tactic better than voting as though the voter did not know anything at all about the stances of the other voters and the popularity of the candidates with them. Such a tactic would effectively take into account only the one voter's affinities toward the candidacies. This would be the naïve tactic. The naïve tactic is obvious for every seriously-proposed voting system. For example, with the choose-one system, this is to vote for ones favorite candidate. Many of us who write about this area of concern erroneously refer to the naïve tactic as "honest" and as not "tactical". Anyway, the logic in favor of preferring voting systems about which none of us seems to have suggested any more effective tactic than the naïve one says that if we can't think of one, neither can our political opponents, so the system is effectively tactic-proof and all one has to do is vote ones affinities.
A concern I think of with this approach is that just because none of us finds a better tactic, does not mean that operatives willing and able to throw a lot of money at the problem won't.
If the concern is valid, a way to address it is, instead of advocating for a system where it is hard to figure out the best tactic, advocate for one for which plausible tactics can be taught and in which if everyone uses a pretty good tactic, they will get approximately equal power over the outcome. For Score Voting, I believe that for a voter whose favorite candidates are not popular and who is facing a situation where one of the front runners is significantly worse than the other one in the estimation of that voter, the tactic is to exaggerate support for the lesser evil, but only within that candidate's rank, not up above a candidate the voter prefers more.
But the drawback of preferring Score Voting over systems where the tactic is hard to understand is that even though the tactic for Score is easier than those for competing systems, it is still too hard to teach to typical Americans. No one knows from probability etc.
So, it's a dilemma, whether to prefer systems where the tactic is almost known, or prefer ones where it is complicated and obscure.
Approval-ordered Llull (letter grades) [10], Score // Llull [9], Score, STAR, Approval, other rated Condorcet [8]; equal-ranked Condorcet [4]; strictly-ranked Condorcet [3]; everything else [0].
-
@jack-waugh said in Ease of Teaching Tactics:
though the tactic for Score is easier than those for competing systems, it is still to hard to teach to typical Americans.
Really? Gotta hard disagree on that.
Say you've got 7 candidates, but only two are the ones that anyone talks about. Those two are well known to be the obvious front runners.
And let's assume score voting has been around for a while. As in, this isn't the first time people are using it, or hearing about it on the news.
Are you saying that "typical Americans" are incapable of figuring out that rating your favorite of those two candidates a maximum score, and your least favorite of the two your minimum score, is going to be the most powerful way to express your vote?
Keep in mind, even if they aren't clever enough to figure that out themselves, people are going to discuss this on the news and elsewhere (at least in any context in which Score was used for real, consequential elections).
So, it's a dilemma, whether to prefer systems where the tactic is almost known, or prefer ones where it is complicated and obscure.
It would be one thing if it was a just a situation of "do you know the clever tactic?" But it isn't like that.
First of all, you not only need to know the tactic, but you also need to have information as to how others feel about the candidates and how they will probably vote. Those are two different things. So I might be a math whiz who thoroughly understands how IRV can best be manipulated. But that won't do me much good if I don't have any ability to guess how others are going to vote with the precision necessary to use this tactic with any effectiveness.
Then there is the question as to how much is to be gained by such manipulation.
With IRV and even moreso with Condorcet-compliant methods, you would typically need high precision information to be able to a move the needle a very very tiny amount. With Score and Approval, low precision information is enough to allow you to move the needle a whole lot more.
Your way of putting it is an extreme oversimplification, that ignores how much is to be gained by attempting to game the system. It's like you are concentrating on how difficult it is to crack a safe, and ignoring the question of how much money is inside the safe.
-
advocate for one for which plausible tactics can be taught and in which if everyone uses a pretty good tactic, they will get approximately equal power over the outcome.
If this is the goal, then you will want IRV with the naive tactic, as all available evidence points towards its difficulty of manipulation. In fact, this is the main strength of IRV. I think I linked this to @rob on reddit as well haha
Single transferable vote resists strategic voting (420 citations)
Social Choice Observed: Five Presidential Elections of the American Psychological Association (97 citations)
The vulnerability of four social choice
functions to coalitional manipulation of preferences (72 citations)Voting rules, manipulability and social
homogeneity (68 citations)Four Condorcet-Hare hybrid methods for single-winner
elections (25 citations)Strategic voting and nomination (20 citations)
Statistical evaluation of voting rules (27 citations)
An Empirical Study of the Manipulability of Single Transferable Voting (71 citations)
On the complexity of manipulating elections (33 citations)
Towards less manipulable voting systems
Strategic, sincere, and heuristic voting under four election rules: an experimental study (101 citations)
On the manipulability of voting systems: application to multi-operator networks (5 citations)
-
@andy-dienes said in Ease of Teaching Tactics:
I think I linked this to @rob on reddit as well haha
I knew that was you!
This one is really good, even if you skim over it: https://hal.inria.fr/tel-03654945/document
Here is the abstract, emphasis mine:
We study coalitional manipulation of voting systems: can a subset of voters, by voting strategically, elect a candidate they all prefer to the candidate who would have won if all voters had voted truthfully?
From a theoretical point of view, we develop a formalism which makes it possible to study all voting systems, whether the ballots are orders of preference on the candidates (ordinal systems), ratings or approval values (cardinal systems), or even more general objects. We show that for almost all classical voting systems, their manipulability can be strictly reduced by adding a preliminary test aiming to elect the Condorcet winner if there is one. For the other voting systems, we define the generalized Condorcification which leads to similar results. Then we define the notion of decomposable culture, an assumption of which the probabilistic independence of voters is a special case. Under this assumption, we prove that, for each voting system, there exists a voting system which is ordinal, shares certain properties with the original voting system, and is at most as manipulable. Thus, the search for a voting system of minimal manipulability (in a class of reasonable systems) can be restricted to those which are ordinal and satisfy the Condorcet criterion.
In order to allow everyone to examine these phenomena in practice, we present SVVAMP, a Python package of our own dedicated to the study of voting systems and their manipulability. Then we use it to compare the coalitional manipulability of various voting systems in several types of cultures, i.e. probabilistic models that generate populations of voters equipped with random preferences. We then complete the analysis with elections from real experiments. Finally, we determine the voting systems with minimal manipulability for very low values of the number of voters and of the number of candidates, and we compare them with the classical voting systems of the literature. In general, we establish that Borda's method, Range voting, and Approval voting are particularly manipulable. Conversely, we show the excellent resistance to manipulation of the system called IRV, also known as STV, and of its variant Condorcet-IRV.
-
Are you saying that "typical Americans" are incapable of figuring out that rating your favorite of those two candidates a maximum score, and your least favorite of the two your minimum score, is going to be the most powerful way to express your vote?
I believe that is not the best strategy for Score if you want a Nader-type candidate to win. I am hoping that the simulator I am working on will confirm or refute ("refudiate") my belief on this, which is that the best strategy is to give slightly less support to the compromise candidate than to the preferred candidate.
And the reason I said "strategy" rather than "tactic" is that even though your preferred candidate might not win the current election, you are still interested in showing support in the totals for that candidate so as to draw the attention of other voters so they check the candidate's stances on the policy issues and that candidate's qualifications and so the candidate might win a future election. So, since your strategy involves raising the reported numbers for how well the candidates you favor did in the election, to improve their chances in future elections, your tactic must aim at those measures, in addition to concerning itself with who is going to win this time.
Keep in mind, even if they aren't clever enough to figure that out themselves, people are going to discuss this on the news and elsewhere (at least in any context in which Score was used for real, consequential elections).
Yes, that works once the system is in place, but while we are still in the stage of advocating for the system, it doesn't help sell the system.
First of all, you not only need to know the tactic, but you also need to have information as to how others feel about the candidates and how they will probably vote.
Of course! That's the subject of the Gibbard theorem. That's why we even need to think about tactics other than the naïve tactic.
I believe that how that figures in in Score is that as your preferred candidate gains in popularity, you lessen your exaggeration of the score you give the compromise.
Then there is the question as to how much is to be gained by such manipulation.
What is to be gained is equal power to that of the other voters.
If the optimal (or a near-optimal) tactic is easy to understand for the system, and the system provides formal balance (along with additivity and expressivity), everyone can use the tactic and thereby have equal power.
If the system provides formal balance but its tactics are hard to understand, then people, factions, and parties that don't have enough money to pay fancy mathematicians to study it for a long time may fail to find an adequate tactic, and so may receive less than equal power, voter for voter.
Approval-ordered Llull (letter grades) [10], Score // Llull [9], Score, STAR, Approval, other rated Condorcet [8]; equal-ranked Condorcet [4]; strictly-ranked Condorcet [3]; everything else [0].
-
@jack-waugh said in Ease of Teaching Tactics:
I believe that is not the best strategy for Score if you want a Nader-type candidate to win.
I would consider a rational person who wants Nader to win, but understands that Gore and Bush each have a a much greater likelihood of winning, and they significantly prefer Gore to Bush --- would be wisest to give Gore maximum rating. Which of course reduces it to Approval (while disadvantaging people who find a sincere vote to be more honorable or ethical, even if not strategic).
If ALL they want is for Nader to win, and they have no other priorities at all, you are right. But I don't see that as a common perspective.
The situation for Score is basically the same as for plurality in that regard. I know a bunch of people who voted for Nader under plurality, for reasons I consider both admirable and unwise, and subsequently regretted it. The same thing would happen under Score.
-
@rob If I finish this friggin' software, we may find out.
